To address the challenges of maintaining solution diversity, insufficient convergence, and inadequate adaptability to complex Pareto fronts in many-objective optimization problems, this paper proposes a grey relation-based evolutionary algorithm with self-adaptive reference vectors. First, to balance the diversity and convergence, a novel environmental selection strategy based on grey relational degree is proposed. This strategy employs grey similitude relational degree to measure distribution differences of solutions in the objective space, and adopts grey closeness relational degree to quantify the spatial distance between solutions and the ideal point. The former maintains diversity, whereas the latter ensures convergence. Then, a reference vector adaptation strategy is developed to track the topological evolution of complex Pareto fronts through dynamic insertion of new vectors and periodic deletion of redundant vectors. Experimental results on 9 DTLZ benchmark problems and two real-world cases demonstrate that the proposed algorithm significantly outperforms four other state-of-the-art algorithms in both convergence and diversity metrics.